See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/272830842

Predicting and Controlling Ground Movements

Around Deep Excavations
CONFERENCE PAPER MAY 2010

CITATION

DOWNLOADS

VIEWS

15

16

3 AUTHORS, INCLUDING:
Malcolm D. Bolton
University of Cambridge
171 PUBLICATIONS 2,656 CITATIONS
SEE PROFILE

Available from: Paul Vardanega

Retrieved on: 17 August 2015

Keynote lecture presented at Geotechnical Challenges in Urban Regeneration,

the 11th International Conference of the DFI-EFFC, London, 26-28 May 2010.

PREDICTING AND CONTROLLING GROUND MOVEMENTS

AROUND DEEP EXCAVATIONS
Malcolm D. Bolton, Sze Yue Lam and Paul J. Vardanega
ABSTRACT
Deep excavations frequently cause problems, and sometimes trigger catastrophic
collapses, especially in soft clay. In principle, these problems are well understood,
but designers may fall between the two stools of naive empiricism and overelaborate finite element analysis (FEA). A new approach, Mobilizable Strength
Design (MSD), has been developed to bridge this gap. MSD specifies deformation
mechanisms tailored to each stage of construction. Each stage is analysed for
energy balance, with incremental subsidence creating a drop of potential energy
which must equal the work done deforming the soil and the support system.
Incremental deformations are summed, while soil non-linearity is allowed for. The
non-linear response of a representative shear stress-strain test is required, but
estimates can be based on routine soil characterisation. It is demonstrated that
MSD back-analyses not only fit FEA results for soft clay within 30%, but also fit
the soil-structure deformation data of 110 field studies within a factor of 1.4. Finally,
a new set of dimensionless groups is defined to characterise deep excavations in
clay without the need for any analysis at all. These are used to chart the maximum
wall displacements taken from the field database, and an elementary formula is
proposed which predicts these 110 maximum displacements within a factor of 2.9.
Guidelines are deduced for designers. In particular, it is shown that wall stiffness
within the typical range of sheet-piles, secant piles and diaphragm walls has little or
no effect on wall deformations.
MOTIVATION
Urban regeneration calls, above all, for rapid
transit links. The raison detre of a city is the
capacity of its citizens to perform transactions in
person. This calls for travel: to work and play, to
buy and sell goods, to offer and access services.
However, the recent rate of growth in car travel is
unsustainable. In order to reduce carbon dioxide
emissions, other pollutants, and noise, car travel
must be replaced by mass rapid transit systems.
Cities around the world are attempting to
regenerate or expand by building subways. At the
same time, property developers are specifying
deep basements which maximise land use. All
these trends call for deep excavations.

brace parallel sides apart, sacrificing some

mobility and workspace inside the excavation. But
for deeper excavations in soft ground the
incremental displacements, which accrue with
every stage of excavation before a new level of
props can be placed, can give rise to significant
bending moments in the walls and to subsidence
affecting neighbouring structures and services.
It is believed that cast insitu reinforced concrete
retaining walls offer some mitigation of these
effects. Cement stabilization can alternatively be
used to create gravity walls. Wherever possible,
the bases of walls are fixed within hard layers
underlying the soft ground. As excavation
proceeds, props or anchors are installed to
prevent excessive inward movement of the walls.

THE STATE OF THE ART

Structures
Excavations can be made by driving deep soldier
piles supporting lateral boards or sheets which
retain the earth. Interlocked steel sheet-piles have
generally superseded them in the absence of
buried obstacles. However, such flexible retaining
systems may bend and bulge excessively when
the depth of excavation begins to mobilise a
significant fraction of the soils available strength.
Props at regular vertical intervals can be used to

Speed of construction is often an issue, and it has

become popular to design the excavation support
system as part of the permanent structure, rather
than as temporary works. A ground floor slab can
be cast as a prop at the crest of the retaining walls
so that the superstructure construction can
proceed above, at the same time as top-down
excavation beneath the ground floor slab permits
the successive casting of basement slabs which
also act as props.

In very deep soft soils it may not be possible to fix

the bases of the walls in stiff ground. It may then
be necessary to use jet-grouting or deep soil
mixing, for example, to introduce a cemented plug
between the bases of the walls prior to
excavation. Such a plug can serve two functions:
as a compression plug acting as a prop, and as a
shear plug preventing base heave.

Secondly, soil will often be partially drained. An

analyst might suggest using a Finite Element
Analysis (FEA) coupling deformation and transient
flow to compute an earth pressure that would
presumably change from an undrained to a
drained value. However, the uncertainty and
variability in ground permeability is even greater
than that for other soil characteristics.

Soil mechanics

Finally, and most significantly, the strains that can

be tolerated around a deep excavation will
generally be much smaller than those required to
mobilise peak soil strength, making parameters cu
and inapplicable. The response in practice is
generally to rely on factors of safety specified in
Codes of Practice, Guides and Regulations. A
more analytical response would be to run an FEA
with appropriate soil deformation parameters. But
this requires exceptionally good judgement in the
selection of mathematical models for soil
behaviour, and in obtaining representative values
of the defining parameters.

The classical soil mechanics literature of earth

retention is expressed in terms of the fully active
and fully passive earth pressures, a and p.
Expressions are developed in terms of MohrCoulomb strength envelopes, described either by
the undrained shear strength cu, or the drained
shear strength . For example, in Rankines
simple analysis for frictionless walls retaining level
ground with a hydrostatic water table:
p

a = z 2c u
p
a

(1)
2

= w zw + ( z w zw ) tan ( 45 / 2)

(2)

where z is the depth below ground level, zw is the

depth below the water table, and and w are the
unit weights of the bulk soil and of the water,
respectively, and refers to passive (+) and active
() pressures for peak soil strengths. Plasticity
theory permits the creation of formulae or tables
which extend these elementary results to rough,
sloping walls retaining sloping ground. Coulombs
wedge analysis can easily be programmed to
compute more complex stratifications and
groundwater variations, but it must be recalled
that assumed mechanisms always provide
solutions somewhat on the unsafe side.
Base heave can similarly be treated with simple
strength parameters in ideal cases. The undrained
shear failure of soft soil can be treated as a
negative bearing capacity failure, with a factor Nc
in the range 5 to 8 as a function of excavation
geometry:

H max = N c c u

(3)

Base heave in the fully drained case will occur

when the passive thrust exerted by the retaining
walls overcomes the vertical effective stresses
beneath the excavation, which will be reduced by
upward seepage.
Significant uncertainties arise in practical
applications, of course. First, the designer must
exercise judgement in order to choose a value of
cu or whilst recognising spatial variability.

At issue is the difficulty that practising engineers

habitually measure the strength of soil, not its
stiffness. It is even common for designers to
eschew proper stress-strain testing altogether in
favour of probing with SPTs, for example. This
cavalier approach to material testing fits very
badly with the trend amongst academic research
workers over the last 30 years to develop ever
more sophisticated constitutive models. These will
usually have been validated against equally
sophisticated batteries of triaxial tests on initially
identical samples of archetypal soils reconstituted
and reconsolidated in the laboratory. But for real,
variable, soils it will be impossible to fix values for
the 20 or 30 parameters that complex constitutive
models usually possess.
The practising engineer who wishes to deal
explicitly with ground strains is currently left
performing FEA with simpler soil models. Since
soil is quite non-linear, the use of linear-elastic
perfectly-plastic soil models is beset by the
problem of selecting an appropriate secant
stiffness through a curve. It would therefore
appear best to attempt to define the stress-strain
curves of representative samples of soil, and to
use these curves directly in FEA. This approach
was first outlined by Duncan and Chang (1970)
who fitted a hyperbola between an initial elastic
stiffness and a peak plastic strength. Reasonable
correspondence has been reported between
measured excavation performance and FEA backanalyses based on sampling, testing and the
judicious fitting of hyperbolae. Nevertheless,
designers are left acutely conscious that the risk
of error in terms of ground deformations is
currently unquantifiable.

Groundwater flow
The
designer
must
also
consider
the
consequences of groundwater pressures giving
rise to flow. Water tends to flow into deep holes,
through the walls and beneath the excavation. If it
is decided to keep the excavation dry, a sump
within the excavation may be sufficient in finegrained soils. But in every case it is necessary to
consider the base heave which may be caused by
deep water pressures. A catastrophic heave
failure will occur when a deep, permeable soil
layer beneath the excavation can maintain a water
pressure as high as the self weight per unit area
of the overlying soil that remains. This is a
particular danger in soil profiles with clay overlying
sand because the contrast in permeability can
maintain high water pressures in the sand as the
clay is excavated. For excavations in coarse
grained soils the heave failure criterion is
equivalent to there being a critical upward
hydraulic gradient which can eliminate all effective
stress beneath the excavation, leading to
liquefaction or piping with internal erosion.
Even if the deep water pressures are smaller than
those required for gross heave, they can create
sufficient softening to severely reduce passive
pressures below the excavation if the soil has time
to swell in response to the removal of overburden.
These threats may lead the engineer to provide
for deep dewatering. However, any removal of
groundwater from the vicinity of an excavation will
tend to lower the external water pressures leading
to subsidence which can affect a wide area. Local
subsidence can be caused by sand production
from badly screened wells or water ejectors.
An alternative solution to dewatering may be to
prop the walls apart at the crest, then flood the
excavation as it is made, ultimately casting a
concrete base slab under water: Clarke and
Prebaharan (1987). If the inevitable construction
difficulties can be overcome, the soil support
system will be complemented by water pressures,
and no transient flow need occur.
RISK
The creation of subways appears to be a
particularly risky business both for the
construction team involved and for the
neighbours. Deep excavations for cut and cover
tunnels, station boxes, and shafts are the subject
of this paper. In the last six years alone there have
been several catastrophic collapses, each with
multiple deaths and injuries, and each posing a
severe commercial threat to the companies
involved in design and construction and a political
embarrassment for the promoters.

These collapses include:

the braced diaphragm wall excavation for
approach tunnels to the Nicoll Highway
subway station in Singapore in March 2004,
the 40m diameter sprayed concrete shaft
construction at Pinheiros Station in Sao
Paolo in January 2007,
the access tunnel to Suzhoujie Station under
construction in Beijing in March 2007,
the braced excavation for the Xianghu
subway station construction in Hangzhou in
November 2008,
the Cologne city archives building falling into
the anchored diaphragm wall excavation for
the construction of Severinstrasse subway
station in March 2009,
and the braced excavation for Sajinqiao
subway station in Xian in August 2009.
When such serious incidents occur, the
emergence of an engineering account of the
probable causes is often hampered by the
commercial, legal and political processes which
generally take precedence. An exception is the
Singapore Governments Committee of Inquiry
(Magnus et al, 2005) into the Nicoll Highway
collapse, which facilitated a review by international
engineering experts, ultimately providing public
access to the findings of serious design flaws. Of
these, two were found to be paramount:
an erroneous characterisation of the soft
Singapore marine clay in computer software,
which led to an overestimation of its
undrained strength by a factor of about 2 and
thereby to an underestimation of the wall
bending moments and the strut forces, albeit
by a much smaller margin;
an under-design of the steel connections
between struts and waler beams, which failed
unstably by buckling at about half their
intended capacity.
The COI put these particular failings into a much
broader context, calling for clearer designation of
responsibilities and a reflective approach to the
identification of hazards and the management of
risks. For example, the COI drew attention to the
large wall deformations (in excess of 400mm) that
had been observed to occur prior to collapse, and
to the potential for monitoring to reveal incorrect
design assumptions long before ultimate
catastrophe. The Singapore Government Building
and Construction Authority (BCA) took immediate
steps in their Advisory Note 1/05 to bring deep
excavations into a clear regulatory framework,
notwithstanding their having previously been
classed as temporary works. This has itself
prompted some vigorous discussion within the
profession, and thereby to amendments in the

recommended procedures which were embodied

in BCA Advisory Note 1/09. Whatever ones views
of the merits of Government regulation, the
professionalism and transparency of the
Singaporean approach should be commended.
In three of the other catastrophic collapses listed
above, press reports speculated that the potential
significance of ground movements may have been
overlooked as construction continued. And the
availability of water from rain storms, broken
pipes, or a nearby river was often linked with
the sequence of events on the day in question. Of
course, ground movements can lead to water
release, as well as the other way round.
Significant ground movements can cause joints to
open up in fresh or foul water services, and
resultant erosion and cracking can lead to the
unexpected arrival of water in the vicinity of the
excavation. Such water may have two deleterious
effects on the stability of earth retaining structures,
an increase of active forces due to the full-height
hydraulic thrust acting on the retained face, and
the removal of passive support due to upward
seepage in the excavation which could even lead
to soil liquefaction, piping or base heave.
In the absence of further information in these
other cases, one can only presume that some
combination of the following shortcomings was
responsible for the catastrophe:
the actual ground profile (such as the depth
of soft layers, their strength, etc) was less
favourable than that assumed in design;
the groundwater regime was less favourable
than that assumed in design;
the surcharge loads actually applied to the
retained surface exceeded those that had
been allowed for;
the sequence of excavation stages in relation
to the placement and pre-loading (if any) of
supports, was more aggressive than had
been allowed for;
the behaviour of one or more of the structural
components of the wall and its supports,
whether props or anchors, was inferior to
what had been assumed in design;
the composite system behaviour of the
ground and the support structure differed
from that assumed in design, to such an
extent that excessive deformations and
displacements
were
induced
during
excavation.
The first five require robust integration between
design, ground investigation and inspection
procedures during construction. It is the sixth, the
creation of a strain-based theory of soil-structure
interaction, which this paper addresses, enabling
a link to be made between factor of safety, wall
deformation, and ground movement.

THE OPPORTUNITY FOR IMPROVEMENT

In their General Report on Deep Excavations, Tan
and Shirlaw (2000) remark that in view of the
uncertainties in ground conditions, analytical
methods, and construction procedures, engineers
generally follow ...a wise course; they build a
sufficiently stiff retaining and bracing structure
which is so strong that the stiffness of soil
contributes little to the overall stiffness of the soil
structure system. If so, these engineers are
treating the soil rather like a heavy fluid into which
a support structure has to be introduced to
maintain a reduced fluid level corresponding to
the excavation, with lateral bracing carrying lateral
earth pressures, and with some means of
preventing the fluid entering from below.
One might therefore compare the construction of a
deep excavation in soft soil with the construction
of a ship top-down, floating in heavy water. Of
course, a marine architect has the advantage of
designing a ship to be built bottom-up, from the
keel, on dry land. So the half-built vessel will be
subject only to gravity while it is being
constructed, not to external soil or water
pressures, and formwork can be used to prevent
deformation. The final geometry of the ship can be
predicted and controlled within tight tolerances, so
the design can incorporate single or double hulls
with prefabricated beams, plate stiffeners and
stanchions all welded in place to create an
efficient integral structure. Clear load paths,
following lines of relative stiffness, enable a
straightforward set of design rules to be used to
produce an initial design that should be safe and
serviceable in the specified loading scenarios of
waves and cargo. The initial design can later be
finessed by finite element analysis to reduce
weight and enhance fatigue resistance, for
example.
The promoters of underground constructions must
regret that their schemes proceed with an order of
magnitude less predictability than the building of a
boat. But they might admire the ingenuity of the in
situ construction process in the light of
comparative costs. A medium-sized oil tanker
might offer useful cargo space of 80 000 m3 for a
construction cost of about 1200 per m3. The
Westminster underground station box for the
Jubilee Line Extension created a similar
underground volume and was tendered at about
1000 per m3 at todays prices, but the
complexities of the site and the need to protect the
adjacent Big Ben clock tower added very
considerably to the final construction cost.
Is further economy possible? In his General
Report, six years after Tan and Shirlaw, Boone

(2006) reflected on whether a less conservative

approach to deep excavation design could be
advocated. In view of the risks and uncertainties
he concluded: Presently, no.. But he did make
the following proposal...
if sufficient testing can be carried out or
collected for specific soil deposits, the
uncertainty (variability) of the engineering
properties may be characterised;
if a sufficient number of Class A predictions
can be made and compared to field
measurements, the uncertainty in analysis
methods may be defined;
if a sufficient number of case histories can be
collected with appropriate identification of
construction details, the uncertainty related to
workmanship and process details may be
characterised;
then it may be possible to complete reasonable
stochastic analyses of excavations so that the
probability of achieving performance goals could
be quantitatively evaluated, designs optimized,
cost-effectiveness improved, and better decisions
made. Boone called for focussed research to that
end.
This paper reports work at Cambridge University
that has been devoted to this challenge.
Responses are given in each of the three areas of
challenge a database of soft clay stiffness, the
creation and validation of a new theory, and the
performance of that theory in back-analysing a
new database of excavation case histories.

stress and strain axes is required. Shear stress, or

mobilised strength, is presented as a proportion of
the peak undrained shear strength, cmob/cu. Shear
strain is normalised in relation to a reference
strain corresponding to the mobilisation of peak
strength, /u. The reference strain u should be
derived by extrapolating a best-fit parabola
through the central segment of stress-strain data
(0.2 < cmob/cu < 0.8) until it reaches peak strength.
Each such parabola is constrained to have its
apex at the origin. The excellent outcome for the
13 soft clays is shown in Figure 1. It is also
evident that a linear lower bound may be advised
at small mobilisations, cmob/cu < 0.2.
BS8002 (1994) defines cu/cmob as the mobilisation
factor M which is equivalent to a factor of safety
on shear strength, though it was explicitly
concerned with limiting deformations rather than
with safety as such. Accordingly, the mobilisation
of shear strength in soft to firm clays (OCR < 5)
can simply and accurately (R2 = 0.98) be written:

/u = 0.2 cmob/cu = 0.2 / M

/u = (cmob/cu)2

= 1/M

for M > 5

(4a)

for M < 5

(4b)

It remains necessary, of course, to select

appropriate values for cu and u in any given case,
ideally through tests on high-quality cores, or
through self-boring pressuremeter tests. However,
a statistical correlation between u and plasticity
index Ip, obtained by Vardanega and Bolton
(2010), and shown in Figure 2:

USING A DATABASE OF SOIL STIFFNESS

u = 0.024(I p ) + 0.0031

Geotechnical
engineers
have
traditionally
approached serviceability and deformation issues
through elastic-style calculations based on some
stiffness modulus. Soil is linear elastic at very
small strains (e.g. shear modulus Gmax at shear
strain < 10-5), but it then suffers a progressive
deterioration of secant stiffness G = /, up to its
peak strength cu. This non-linearity is generally
expressed through the normalisation G/Gmax = f().
However, it will be useful here to characterise the
mobilised shear stress = cmob (< cu) = G = f*().
We can then deduce the shear strain that would
be experienced at some mobilised shear stress
cmob smaller than the peak undrained shear
strength cu.

permits the engineer to select a reasonable value

for lightly overconsolidated clays and silts.

This mobilisation of strength approach has been

used by Vardanega and Bolton (2010) to present
the data of 13 soft clays and silts previously
published by Anderson and Richart (1976), Kim
and Novak (1981) and Shibuya and Mitachi
(1994). For this purpose, normalisation of both

(R2 = 0.81)

(5)

In applying equations 4 and 5 it should be noted

that they were derived using the data of cyclic
tests. It was implicitly assumed that strain rate
would have effected cmob and cu values equally,
and would have cancelled out. Ideally, an
engineer should perform a stress-strain test on a
representative sample at a strain rate appropriate
to the design situation. However, the errors
incurred by applying a simple rule such as 5%
strength enhancement on cu per factor ten
increase in strain rate, following Vucetic and
Tabata (2003), might be considered acceptable.
Equation 4 could then be applied with a ratecorrected strength cu* and the reference strain u
obtained from equation 5 in order to estimate soil
strains in service. In some circumstances, a
designer might even be prepared to infer cu from
an SPT blowcount, following Stroud (1974), in
order to make a preliminary assessment of ground
strains using (4) and (5).

1.2
1

cmob/cu

0.8
0.6
0.4
0.2
0
0

0.2

0.4

0.6

0.8

1.2

Normalised Strain (/u)

Figure 1 Normalisation of shear stress-strain

curves from cyclic torsional tests on clays

Reference Strain (u)

0.07
0.06
0.05
0.04
0.03
0.02

Finally, it is necessary to review what has been

omitted from this section. Firstly, overconsolidated
clays may be found to have a smaller reference
strain u than the same clay when it is normally
consolidated. Secondly, the strength mobilisation
of granular soils has not been addressed here.
And thirdly, no mention has been made of the
volume changes which may follow or accompany
shear. Work is in hand in Cambridge on each of
these aspects with a view to deriving simple
design rules for the limitation of soil strains in
these cases also.

0.01
0
0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

Plasticity Index (I P )

Figure 2 Dependence of reference strain u on

soil plasticity index Ip

Mobilization of undrained shear strength

However, a point of caution should be noted

regarding
the
simplified
parabolic-to-peak
mobilisation curve of Figure 1. Monotonic triaxial
tests on undisturbed samples of soft clay are
sometimes found to give strains in excess of the
parabolic fitting for cmob/cu > 0.8. See Figure 3,
taken from Lam and Bolton (2010), for examples
of shear stress-strain mobilisation curves
associated with field records of deep excavations
in soft clay, for example. Furthermore, most
normally consolidated clays are sensitive, so their
available strength reduces with increasing strain
after the peak. It would therefore be prudent not to
permit a mobilisation factor M < 1.2 for design
purposes unless the retaining wall is fixed in a
hard layer at its base, and is sufficiently stiff to
control soil strains.

1.2

u=1.0%

u=5.0%

u=3.0%

1.0
0.8
0.6

Mexico City Clay (Diaz-Rodriguez et al.,1992)

Bangkok Clay (MOH et al., 1969)
Oslo Clay (Bjerrum and Landva, 1966)
Boston Blue Clay (Whittle, 1993)
San Francisco Bay Mud (Hunt et al., 2002)
Chicago Glacial clay (Finno and Chung, 1990)
Shanghai Clay (X.F.Ma,Person. com., Feb 2009)
Taipei Silty Clay (Lin and Wang,1998).
Singapore marine clay (Wong and Brom, 1989)

0.4
0.2
0.0
0

10

Shear Strain, (%)

Figure 3 Mobilisation curves for various case

histories of deep excavation

GROUND DEFORMATION MECHANISMS

Figure 4 shows various simplified deformation
mechanisms applicable to retaining structures
supported in different ways. In each case,
deformations in the shaded zone are ignored in
favour of a specified zone of finite shear strains
giving rise to ground and wall deformations that
are consistent with the boundary conditions.
These mechanisms require that the soil deforms
at constant volume undrained in the context of
clays deforming during excavation. They can be
used to produce approximate solutions for ground
deformations in a fashion analogous to the use of
slip circles or wedges in collapse limit analysis.
And, in the spirit of upper bound limit analyses,
they tend to overestimate the depth of excavation
that will generate a given volume of subsidence.
Just as with limiting equilibrium solutions, these
limiting deformation solutions therefore need to
be checked against FEA, centrifuge models, or
field records to ensure they are adequate.

relation = 2, derived from Mohrs circle of strain.

Centrifuge model tests showed that this approach
can provide reasonable accuracy, notwithstanding
the simplifications of focusing only on near-field
soil deformations (see Figure 4a) and ignoring the
influence of wall roughness.
a) stiff wall pinned at its base in a hard layer

b) flexible wall fixed at its base in a hard layer

c) stiff wall propped at its top

H
D

d) flexible wall bulging below fixed props

Figure 4 Simplified deformation mechanisms

MOBILIZABLE STRENGTH DESIGN (MSD)

The first application of MSD was to rigid, smooth,
cantilever retaining walls in clay: Bolton and
Powrie (1988). The wall was supposed to respond
to excavation by rotating about a point just above
its toe, creating active and passive zones either
side, but restricting the strength in equation 1 to a
value cmob < cu. The equations of lateral and
moment equilibrium were written down and solved
for cmob. The mobilised shear strain could then
be read off a representative stress-strain curve,
and the wall rotation could be deduced from the

A follow-up study by Osman and Bolton (2004)

compared MSD based on wall rotation with FEA
for floating cantilever walls, and showed that the
influence of Ko 1 could be accounted for if the
data of compression and extension tests were
separately available. Neither wall friction nor the
precise shape of a soils stress strain curve
caused MSD calculations to depart excessively
from FEA computations. However, one factor not
included in the MSD wall rotation model was
significant in the FEA: the base of the wall
translated towards the excavation. Effectively,
FEA showed that a floating wall analysis
invoking only wall rotation (Figure 4a) caused
MSD to underestimate wall crest displacements
by up to a factor of 2. MSD needs to include wall
bending (Figure 4b), and deep foundation
shearing (Figure 4c), if more precise predictions
are to be made.
These aspects were considered by Osman and
Bolton (2005) in the application of MSD to braced
excavations, which specifically addressed the
bulging that is seen below the bottom row of props
when the excavation is advanced. Each increment
of bulging can be predicted reasonably well by
assuming the bulge to be sinusoidal, following
ORourke (1993). Figure 4d shows a wall fixed in
a hard layer, for which the wavelength of the bulge
at any stage of excavation equals the distance
between the bottom row of props and the bottom
of the soft clay. Where there is no hard layer,
virtual fixity can be imagined at a depth below the
wall base equal to half the unsupported wall
length below the bottom prop. Wall displacement
calculations begin with the cantilever stage
(Figure 4a) as an excavation is made for the top
level of props. Progressive displacements are
found by superposing each stage of wall bulging
(Figure 4d) as excavation proceeds below the
most recently fixed row of struts. Calculated wall
displacements fell within 30% of field records.
These predictions of flexural deformation required
a different calculation method based on the
conservation of energy within the deformation
mechanism, as set out by Bolton et al (2008). The
loss of potential energy, represented by the
formation of a subsidence trough in the retained
ground while the soil within the excavation
heaves, must be balanced by the work done
deforming both the soil and the retaining structure.

The loss of potential energy during any given

stage of excavation is quite easily calculated as:

P = (g z) A

cu

(6)

parabola

where the weight g A of each element of soil of

area A within the mechanism is associated with
its average settlement z. Each settlement z
must be deduced from the sinusoidal
displacement profile flowing through the
mechanism in that stage.
The work done on the soil is calculated using the
swept area under the soils stress-strain curve.
For example, Figure 5a demonstrates that the
work done per unit area in straining soil by amount
along a parabola (equation 4a) is:

Wsoil =

2
2
= cu
3
3 u

0 .5

2 c u 1. 5

3 0. 5

The work done in bending the wall from its initially

undeformed state is calculated from the area
under the moment-curvature relation. Figure 5b
shows that for a wall segment of unit length with a
flexural stiffness EI developing a curvature :

Wwall =

2
1
1
M = EI
2
2

(9)

The magnitude of curvature has to be calculated

from the second differential of the sinusoidal wall
deflections. Then the total flexural work done on
the wall is found by integrating over the length of
the wall:

Wwall = Wwall dx

(10)

The fundamental energy conservation equation of

MSD is then, simply:

P = Wsoil + Wwall

(11)

This can be solved iteratively for each stage of

excavation, using the maximum wall displacement
wmax in that stage as the unknown. The
subsidence increment is the same. Finally, these
increments can be superposed to obtain the
overall maximum displacement wmax.

0. 5

a) Work done per unit volume of soil

flexure
M = EI

(8)

Wsoil

(7)

The magnitude of strain has to be calculated

from the sinusoidal displacement field. Then the
total work done on the soil, starting from an
undeformed state, is found by integration over the
whole area of the mechanism:

Wsoil = Wsoil A

=
cu u

Wwall

b) Work done per unit length of wall

Figure 5 Unit work calculations
Bolton et al (2008) took care to recognise the
incremental and non-linear nature of this energy
balance in the calculations they advanced. It is
necessary to store the previous shear strain of
each element of soil so that the work done due to
the next increment of excavation can properly be
calculated with respect to the non-linear stressstrain curve. A similar sort of procedure has to be
used for the structural strain energy since this is
also non-linear, having a quadratic relation to wall
displacement and curvature.
Other important details were addressed, including
the need to use a different mechanism when the
excavation is narrow compared with its depth, and
ensuring that the ground model could incorporate
layers with different soil properties. In this fashion,
they set out to compare MSD predictions of
braced excavation performance with those of FEA
conducted by Jen (1998) for a floating wall, with
the parameters recorded in Figure 6. Typical
bulging wall profiles from Bolton et al (2008) are
compared in Figure 7 for a 17.5 m deep
excavation supported at 2.5 m intervals, against a
40 m deep wall in a 100 m deep profile modelled
as Boston Blue Clay. Excavation width B is varied
from 30 m to 60 m.

Retaining wall
(EI=1440, 77 and 19MNm /m)
s=2.5m
H
L=20, 25, 30, 35 and 40m
C=37.5, 50 and 100m

C
L
B/2=15, 20, 25 & 30m
OCR=1

Although MSD, working in a matter of minutes on

a spreadsheet, does not precisely accord with
Jens FEA, the correspondence is close enough
for decision-making purposes. The two prerequisites for success are to have a compendium
of simplified deformation mechanisms for
compatible soil-structure interaction which are
sufficiently close to reality (or to a good computergenerated simulation of reality), and to be able to
characterise the ground in relation to its density,
strength profile and stress-strain relation. If these
aspects are well-represented the MSD analysis
can hardly go wrong, because the principle of
conservation of energy can hardly be faulted.

Figure 6 FEA carried out by Jen (1998)

A NEW DATABASE OF EXCAVATIONS IN
SOFT CLAY

0
B=60
B=50
B=40
B=30

Depth (m)

10

With the aim of gaining a better understanding of

ground movements around deep excavations, a
database of 155 case histories was collected from
geotechnical journals, international conference
proceedings, national technical reports, and
dissertations. Studies from nine cities are
included: Bangkok (2 sites), Boston (5), Chicago
(10), Mexico City (1), Oslo (9), San Francisco (4),
Shanghai (67), Singapore (21) and Taipei (36).
Full details will be found in Lam (2010).

20
30
40
50
0.00

0.04

0.08

0.12

0.16

Wall deflection, w, (m)

(a) Prediction by MSD

0
B=60
B=50
B=40
B=30

Depth (m)

10
20
30
40
50
0.00

0.04

0.08

0.12

0.16

Wall deflection, w, (m)

(b) Prediction by MIT E3 model (After Jen,1998)

Figure 7 Wall deflections for various widths B

These cases focused on the deformation of

braced walls supporting deep excavations in soft
to firm clays (i.e. cu < 75kPa). For each case
history, information was extracted and analyzed
regarding major factors such as soil properties,
groundwater conditions, excavation geometry,
stiffness of structural support system, construction
method and soil type. A total of 110 out of these
155 case histories were so fully documented that
an analysis could be conducted with the MSD
method taking careful consideration of ground
stratification, the undrained shear strength profile
of the soft to firm clay, its stress-strain behaviour,
and the elevation of the stiff base stratum within
which deformations might be ignored.
Figure 3 showed the mobilization of shear
strength versus strain in tests reported for the nine
soils concerned. The case-specific information
from the database was used directly in MSD to
predict maximum wall displacements wmax for
each stage of excavation, which are compared
with authors reported observations in Figure 8.
Results show a correlation coefficient R2 = 0.91
about the line 1:1. More than 90% of the MSD
predictions fell within a factor of 1.4 of the
corresponding observations. Considering the lack
of any detailed account of soil anisotropy, the field
performance of MSD may be considered quite
satisfactory.

200

R2=0.91
COV=0.25
wmax,m/wmax,p=0.7

150
Predicted w max,p

wmax,m/wmax,p=1

average = D 0.5H

100

wmax,m/wmax,p=1.4

(14)

referring to Figure 4d, and we will use this value in

equation 12.
Now consider the various system parameters that
might modify the structural response ratio. It is
proposed that they can best be visualised in
combination as a soil-structure stiffness ratio R:

50

R=

50

100

150

200

Measured wmax,m

Figure 8 MSD calculations versus measurements

USING DIMENSIONLESS GROUPS
Whilst MSD can confidently be used to make sitespecific predictions, it will also be helpful to use
MSD concepts to derive dimensionless groups to
chart data from field case studies, and to assist
decision-making prior to any detailed analysis.
Consider first the normalization of maximum wall
displacement wmax itself. MSD shows us, as in
Figure 3, that the initial excavation to fix the top
row of props generally produces relatively small
movements at the crest in the cantilever phase of
wall rotation or bending (Figure 4a,b). The most
significant deformations for a fixed-base wall arise
from later stages of construction when the wall
bulges below the lowest props (Figure 4d). Elastic
beam theory applied to the mechanism then
teaches us that the change of net pressure
necessary to produce a bulge of amplitude wmax
over a wavelength will be

p w max

EI

w max EI

gH

G=

Whilst MSD permits to reduce progressively as

props are placed deeper, a design chart might

(15)

c mob

c
= u
u

0 .5

cu u

0. 5

(16)

Using equation 4b again, we can write this as:

G=

cu cu

u c mob

cu

(17)

representing the shear modulus increasing at

higher values of M and correspondingly smaller
strains. Now, the average mobilisation factor M in
the soil should be of the following form:

(12)

(13)

cu cu
u gH EI

where is again taken as the average value

during the whole excavation process, and where
the value of cu is selected to represent the
strength at mid-depth of the soft clay layer. The
rationalisation is as follows. The ratio cu/u is the
secant shear modulus just as peak strength is
reached. For a parabolic shear stress-strain curve
(equation 4b) we obtain the secant modulus
mobilised at any earlier stage:

This can be normalised with respect to the

reduction in vertical earth pressure gH due to
excavation. Accordingly, a structural response
ratio S can be defined:
S=

simply characterise the average situation, and an

average value will be taken here as the distance
between the middle props and the base of the
deforming clay. We will define

cu
gH

(18)

governing the factor of safety of an undrained

excavation. It follows from (17) and (18) that the
mobilised soil secant modulus:
G

cu cu
u gH

(19)

which is recognisable as the first two terms on the

right hand side of equation 15. The final term
(3 /EI) is simply the inverse of wall bending
stiffness per unit length, expressed in consistent
units. Equation 15 therefore represents the non-

log10S = 2.6 log10R

(20)

and rationalised to give:

1
400

(21)

Substituting for S and R from equations 13 and 15

respectively, we can define a normalised
displacement factor such that:

= SR =

w max 1 c u

u gH

1
400

(22)

Notably, the wall bending stiffness has now

cancelled out. Figures 9 and 10, together with
equations 21 and 22, have shown that wall
stiffness has a negligible influence on the
amplitude of wall bulging due to excavation. This
has been proved for EI ranging within 4 orders of
magnitude between sheet-piles and reinforced
concrete diaphragms, representing 110 braced
retaining walls from around the world.
Equation 22 enables the engineer to estimate
maximum
wall
displacement
wmax
using
knowledge only of excavation depth H, clay depth
D to a stiff stratum (to calculate from equation
14), soil density and shear strength cu at middepth 0.5D, and the reference strain u required
for peak strength which can be estimated from
plasticity index according to Figure 2. But the
possible variation by a factor of up to 2.9 must not
be forgotten. This is emphasised in Figure 11
where the normalized displacement factor is
plotted on a linear scale against a nondimensional structural system stiffness where:
EI
=
(23)
4
w g

R2 = 0.9642

10-1
10-2
10-3
S=

10-4
R=

wmax EI

gH4

cu

cu 3
gH EI

10-5
0.001 0.01 0.1 1 10 100 1000
Soil-structure stiffness ratio, R

Figure 9 Field data plotted on log S versus log R

100
Structural response ratio, S

within an uncertainty factor of 2.9 (calculated as 2

standard deviations) according to Figure 9. The
same relationship emerges from the predicted wall
displacements using MSD: see Figure 10.

MSD predictions
Field records

10-1
10-2
10-3
S=

10-4
R=

wmax EI

gH4

cu

cu 3
gH EI

10-5
0.001 0.01 0.1

10 100 1000

Soil-structure stiffness ratio, R

Figure 10 Field data and MSD predictions plotted
on log S versus log R

Normalized displacement factor

SR

100
Structural response ratio, S

dimensional soil-structure stiffness ratio for soil

with a parabolic stress-strain curve up to (cu, u).
The new database can now be used to investigate
the relationship between structural response ratio
S and soil-structure stiffness ratio R: see Figure 9
where this is shown on log-log axes to capture the
enormous range of wall stiffness between sheetpiles and thick diaphragm walls. Very remarkably,
the field data fit an exactly inverse relationship
(i.e. the slope equals -1.00), with a coefficient of
determination of 0.964. This can be written as

0.010
Field records
Linear regression line

0.008
0.006
0.004
0.002
0.000
0.001 0.01 0.1 1 10 100 1000
Structural system stiffness

Figure 11 Field data plotted on versus

The arithmetical average value of for the field

data plotted in Figure 11 is 0.003, but the range is
about 0.001 to 0.008. It must be recognised that
much of this scatter arises from the desire in these
charts to represent the excavation construction
process in terms of a single mechanism of
dimension . If the engineer is prepared to take
another few hours establishing the possible
construction sequence and excavation stages,
and performing a cumulative MSD calculation on a
spreadsheet, then the scatter of 110 jobs reduces
from a factor of about 2.9 (see Figure 9) to a
factor of about 1.4 (see Figure 8). Most of this
remaining uncertainty is thought to arise from the
difficulty of selecting representative values of
undrained shear strength cu and reference strain
u. It would be easy to accept that cu might be
uncertain within 10% so that cu2 varies within a
factor of 1.2. And while the plots in Figures 8 and
9 benefitted from an estimate of u based on the
data of site-specific direct simple shear (DSS)
tests, shown in Figure 3, the uncertainty factor
must again have been about 1.2. A larger
uncertainty would have had to be accepted if u
had been estimated from plasticity index using
Figure 2, of course.
DISCUSSION
It is interesting to use equation 22 to develop an
expression for the more familiar term wmax/H:

gH
w max

u
H
400 H c u

(24)

Normally consolidated clays

Let us now estimate the undrained strength at
mid-depth of soft clay. Considering normally
consolidated high plasticity clay with a high water
table:

cu = 0.3 v = 0.3 ( w) gD/2

(25)

Taking the corresponding density 1500 kg/m3,

and making the necessary substitutions:

H
w max
H (D 0.5H )
u
400 u
2
2
H
400 H D
D
2

w max
H
H
u 1 0.5
H
D
D

(26)

For example, if H/D = 2/3 such that = H, and u =

3%, we find:

w max
0.44 u 1.3%
H

(27)

Bolton et al (2008) showed that the average shear

strain mobilised in the soil around a bulging wall is

average 2

w max

(28)

for relatively wide excavations (B ), and

somewhat larger in the passive zone between the
walls for narrow excavations (B < ). Combining
(24) and (28), we find

average

u gH

200 c u

(29)

Using the same wall geometry as in (26) and (27)

we find that for typical braced excavations in
normally consolidated high-plasticity clay:

average

u
200

400

H
D

2
2

0.89 u

(30)

However, shear strains vary within the assumed

deformation mechanism up to a maximum which
is /2 times the average. And there must
ultimately be a tendency to develop a thin slip
zone adjacent to the wall. These considerations
must lead us to expect first a loss of secant
stiffness, and then a drop in strength below the
peak, as large deformations develop behind a wall
in normally consolidated clay. The rate of wall
bulging and ground subsidence would then start to
exceed the initial predictions of equation 24, and
might result in structural failure. And although the
foregoing calculations were predicated on
undrained behaviour, the tendency of soil beneath
the excavation to swell and soften due to stress
relief would compound the problem. It should be
considered essential to monitor the wall profile
continuously during the excavation of normally
consolidated clays.
Overconsolidated clays
If the strength of the clay at its mid-depth was
double the normally consolidated value, equation
24 suggests that the wall displacements would
correspondingly be reduced by a factor of 4. On
the other hand, overconsolidated clays are
heavier, and if the bulk density was 25% higher,
the wall displacements would be 25% larger. It is
not yet clear whether overconsolidation reduces
the reference strain u, so this will be regarded as
being unchanged. The estimated maximum wall
displacement in this firm clay, in the same

excavation scenario as that examined for soft

clay in (25) to (27), would therefore be:
w max
0.14 u 0.4%
H

(31)

The useful notion of parabolic stress-strain curves,

and the application of dimensional analysis, has
allowed us to estimate that wall displacements in
firm clays could well be about one third those of
the same type of soil, had it been normally
consolidated. This offers a straightforward
explanation for why it is normally consolidated
clays that have caused greatest anxiety, even
where braced reinforced concrete diaphragm
walls have been used to support the sides.
The retaining wall
These bulging displacements must apparently be
regarded as unavoidable, even if a typical
diaphragm wall is used. The evidence of Figure 9
suggests that a secant-pile wall or a sheet-pile
wall should fare just as well if it were correctly
braced or anchored. Measures such as
exceptional wall thickness, barrettes or buttresses
would be needed to enhance the structural
stiffness beyond that of any of the walls in the new
database (R < 0.01; > 100) if bulging was to be
suppressed. Only then might the structural strain
energy increase to the point where Wwall
becomes significant compared with Wsoil in
equation 11, and only then might the trend-line of
structural response in Figures 9 and 10 curve
below the linear regression line established for
walls in the existing database. However, the
current study does raise the question of whether
heavy duty diaphragm walls are necessary or
efficient, especially in overconsolidated soils.
An alternative structural design philosophy in stiff
clays would be to accept the wall displacements
and curvatures implied by equation 24, and to
create a well-integrated structure with a wall that
was as thin as possible, to minimise bending
strains. Of course, other constraints must be met,
such as ease of construction, connection to
supports, water-tightness and the walls possible
role in vertical load-bearing.
The supports
If the magnitude of bulging displacements
estimated in equation 22 pose unacceptable
difficulties either for the retaining structure itself or
for nearby structures or services, there are
relatively few responses that offer a viable
remedy. Before considering them, however, it
would be wise to establish the situation with
greater precision. It should be recalled that
equations 22 to 30 derive from the regression in
Figure 9 which is subject to an uncertainty factor
of 2.9 (at two standard deviations) compared to an

uncertainty factor of 1.4 when using incremental

MSD. Most of this difference will arise from
differences in the bracing regime that are not
reflected in the choice of a single representative
wavelength , as used in equation 22. It follows
that by adopting very stringent propping
procedures, with close spacing and pre-stressing,
an engineer might well be able to reduce the
magnitude of bulging by a factor of 2 compared
with equation 22. A slack procedure of installing
widely-spaced struts could, equally, lead to
displacements double those indicated in equation
22. These possibilities could be foreseen by
performing appropriate MSD analyses. If further
assurance were needed, FEA could be performed
using stress-strain data obtained from tests on
high quality cores. The risks of adopting
inappropriate parameters would then be much
reduced if the FE program could accept raw
stress-strain data normalised in the fashion of
Figure 1, rather than attempting to fit a more
general constitutive model. The responsibility for
commissioning appropriate stress-strain tests (e.g.
DSS tests) on representative high-quality samples
would remain with the designer.
If these more rigorous calculation procedures fail
to assure the designer that structural damage
could be avoided using a conventional wall with
high quality bracing, then additional methods of
support will have to be considered. Some of them
are considered below.
As mentioned earlier, the excavation could be
carried out under water. Archimedes principle
could be invoked, with the bulk density in
equation 22 being replaced by ( w), which may
be only one third of its previous value. When
squared, as indicated, this would offer a reduction
factor of 9 in the maximum wall displacement.
Another possibility is to pre-construct deep in situ
props between the walls, prior to excavation.
Access shafts could be sunk at intervals between
the walls to enable precast concrete props to be
installed in sections by jacking. In situ props could
be designed to limit wall bulging to any required
extent. Something similar can be achieved by
introducing zones of jet grouting, deep mixing, or
some other method of ground improvement, to
prop apart in situ walls once they have been
introduced to their full depth: Shirlaw et al (2006).
However, it would be dangerous to place such
props in the expectation of easily removing them
again during the excavation process. If they were
effective as props during the excavation process,
they will have developed strong compressive
reactions. Extremely stringent conditions would
need to be set for the casting of their permanent
replacements prior to their removal, therefore.

One example of a permanent prop created by

ground improvement is a base plug or slab of
stabilized soil introduced by jet grouting or deep
mixing, placed between the side walls, and just
above their base. Plugs have the ability to prevent
wall base translation, thereby converting a
floating wall (Figure 4c) into a fixed-base wall
(Figure 4d). This will have the effect of reducing
both the amplitude of wall displacement wmax, and
its lateral extent . Such plugs are generally more
than 2 m deep to ensure that soft spots due to
poor mixing do not compromise the integrity of the
support system. Particular care has to be taken to
avoid soft clay remaining at the interface between
the wall and the plug, since that would
compromise both the axial stiffness of the plug as
a prop and its shear resistance, which is required
to resist soil and water pressures from below
when the excavation has been completed.
From the standpoint of construction efficiency and
cost, the question often arises of the optimization
of temporary propping within the excavation.
There are some significant difficulties.
The pre-existing earth pressures will be
difficult to assess. Only normally consolidated
soils will have earth pressure coefficient
Ko,nc 1 sin; overconsolidated soils will
possess a higher value.
The construction of a cast-in-situ wall alters
the earth pressure, bringing it close to the
pressure of wet concrete (but not exactly, due
to 3D construction effects): Richards et al
(2006).
The first stage of excavation, prior to the
fixing of any struts or anchors, induces
cantilever rotation or bending (Figures 4a, 4b)
which mobilises some soil strength, cmob or
mob, depending on the permitted strains.
Whilst doing so, the major principal stress
remains nearly vertical, so that lateral
pressures outside the excavation drop to
mobilised active values, in which cmob
replaces cu in equation (1) for example:
Bolton and Powrie (1988).
The subsequent placement of props, with
excavation taking place below, leads to wall
bulging and the further mobilisation of soil
strength, but no longer in a simple active
condition. The shearing in the upper region of
soil behind the props will be on vertical
planes, so the prop forces have no reason to
reduce. Indeed, as the earth pressure behind
the deeper, unpropped portion of the wall
reduces, overall horizontal equilibrium may
require that propping forces in the upper
portions increase.
On the other hand, various factors such as
non-rigidity of props, delays in their

installation, or the introduction of gaps on

purpose, can lead to a further phase of
outward wall movement at the top, with stress
relaxation leading to smaller prop forces.
Temperature changes after installation can
lead to significant changes in prop force, and
to both cyclic and permanent wall
movements: e.g. Batten and Powrie (2000).
Continuity and ductility of props and their
connections is very important, but difficult to
assure because of the tendency for plastic
buckling: Williams and Waite (1993).
The accidental failure or purposeful removal
of a prop or props must lead to load
redistribution, and can engender progressive
collapse unless sufficient redundancy has
been designed-in:.Gaba et al (2003).
In these circumstances it is understandable that
designers generally resort to recommendations for
prop load distributions based on empirical
observations: Terzaghi and Peck (1948), Twine
and Roscoe (1999). Optimization must be seen in
the context of monitoring construction and designas-you-go.
Limits on wmax/
Equation 22 (rewritten as equation 24) sets an
average expectation for the bulging wmax of a
braced wall in terms of the average deformation
wavelength = D 0.5H down to its fixed base at
depth D, during the creation of an excavation of
depth H within a soft to firm clay stratum. The key
parameters are the mid-depth undrained shear
strength cu and the shear strain u required to
mobilise it. This estimate can be refined for
greater accuracy by using MSD incrementally with
actual (intended) prop spacings and excavation
phases, together with the actual soil strength
profile.
The acceptability of the wmax value in any
particular application depends on three criteria:
1. the avoidance of serious damage within the
retaining wall itself;
2. the avoidance of serious damage within preexisting structures or services nearby;
3. the ability to control the safety of the
construction
process
by
monitoring
deformations and introducing additional
supports if necessary.
Regarding the first criterion, it is easily shown that
the maximum tensile strain at a sinusoidal bulge
of magnitude wmax and wavelength in a wall of
thickness d is given by:

max =

w max d

(32)

In the case of a 0.8 m thick diaphragm wall

bulging 0.2 m over an average 20 m wavelength,
for example, (31) gives max 2 x 10-3 assuming
that the neutral axis of bending remains roughly
central. Tensile cracking begins in concrete at
crack 10-4, longitudinal steel reinforcement yields
at about yield 1.5 x 10-3 and concrete crushing
typically begins in compression at crush 4 x 10-3:
Park and Gamble (2000). So the structural
designer may, with care, be able to design the
wall as an under-reinforced, fully cracked section
which maintains ductility at its ultimate moment of
resistance as the wall bulges. A thorough
incremental analysis with MSD would provide a
more realistic wall profile from which a better
estimate of curvature could be made. But it is
argued here that it is the continuity of the wall that
is paramount during excavation, recalling that
equation 22, derived from the case studies,
demonstrated the inconsequentiality of wall
flexural stiffness to its own deflected profile. The
role of the wall is to span between close-spaced
props and to prevent local shear failure beneath
the props.
Burland and Wroth (1974) showed that the
potential damage to structures with shallow
foundations can best be assessed in relation to
the relative settlement /L, defined as the
maximum offset from an initial chord length L.
This can be equated here to wmax/. Burland et al
(2001) explain in detail how to apply this concept
to the assessment of damage accompanying
tunnelling settlement. It is sufficient here to note
that the maximum tensile strain induced in a
building is roughly equal to 1.3/L, and that
severe damage is likely to occur beyond a tensile
strain of 0.3%. Accordingly, a subsidence trough
with relative settlement wmax/ > 0.2% is likely to
cause severe damage. It is instructive to compare
this with the estimate made in equation 30 for the
bulging of a typical wall supporting an excavation
in firm clay, wmax/ 0.14u. This could be no more
than 0.14% in a low plasticity clay, or as high as
0.7% in a high-plasticity clay, as indicated in
Figure 3. It will often be necessary to demolish
and rebuild buildings adjacent to deep
excavations in soft to firm clay, unless measures
(additional to conventional bracing) are used to
suppress bulging. Because the whole depth D of
clay is initially involved in the cantilever phase of
wall movement prior to placing any props, see
Figure 4a, structures and services within a
distance D from the wall must be regarded as
most vulnerable, and meriting detailed analysis.
The third criterion which may place a limit on wall
displacements is the controllability of the
excavation
process
through
monitoring.
Uncertainty over the selection of a representative

value for cu from ground information prior to

construction is compounded by the influence of
localisation,
softening
and
creep
during
construction all of which are exacerbated as
peak strength is approached. This suggests the
desirability of providing a margin of safety (or a
mobilisation factor M) of at least 1.2 against the
best estimate of undrained peak strength.
Assuming a parabolic undrained mobilisation
curve, equation 4b, this corresponds to a
restriction 0.69u. Equation 28 translates this
into a suggested restriction on maximum wall
movement:
w max

0.35 u

(33)

Whereas this might permit wmax/ 1.4% in a

compliant plastic clay such as Singapore Marine
Clay which has u 4%, it would set a stricter limit
of wmax/ 0.4% in Boston Blue Clay with
u 1%, referring to Figure 3.
Equation 33 represents an attempt to ensure that
the rate of displacement with ongoing excavation
remains manageable. Construction monitoring,
which should be regarded as essential, should
then be able to keep pace with events and permit
sufficient time to recognise the approach of failure
so that additional support can be provided.
CONCLUSIONS
1. A new database of clay stiffness has been
used to demonstrate that the undrained stressstrain curve of soft to firm clays can be regarded
for practical purposes as a parabola with its apex
at the origin, passing through the point (cu, u)
where u is found to be related to plasticity index.
2. An incremental Mobilizable Strength Design
method for deep excavations has been formulated
in which conservation of energy is applied stage
by stage to a repertoire of assumed deformation
mechanisms. Deformations are accumulated, but
account is taken of soil non-linearity. MSD
predictions were compared with the results of
Finite Element Analysis previously carried out by
Jen (1998); discrepancies are smaller than 30%.
3. A new database of 155 field case studies of
excavations in soft to firm clays has been
compiled from 9 cities worldwide. Of these, 110
studies are sufficiently detailed to permit MSD
back analyses for comparison with actual
measurements of maximum wall displacement. It
is shown that 90% of MSD predictions fall within a
factor of 1.4 of the actual measurements.

4. New dimensionless groups have been defined

using concepts arising from MSD analysis. These
are based on the typical deformation mechanism,
taken to be that applying when the excavation is
50% constructed. A soil-structure stiffness ratio R
accounts for parabolic non-linearity of the soil. A
new structural response ratio S normalises the
lateral pressure required to create a given wall
bulge by dividing it by the total vertical earth
pressure removed in the excavation. Both R and S
involve the flexural stiffness of the wall, EI.
5. When log10S was plotted against log10R for the
field data, the linear regression had a slope of -1,
with a high coefficient of determination
(R2 = 0.964).
This
indicates
an
inverse
relationship, proving that RS is a constant, and
that EI cancels out. A new normalised
displacement factor = RS is defined with
regression offering a value of 1/400. This permits
deformations to be predicted within a factor of 2.9.
The additional influence of staged construction
and prop spacing is inherent to MSD, explaining
its better performance within a factor of 1.4. One
striking outcome is the proven irrelevance of wall
stiffness within the range from sheet-piles to
diaphragm walls. The wall projecting below the
bottom level of props behaves as flexible in
relation to the adjacent soil, at least in the 110
cases that have been examined.
6. Relative wall deformation wmax/H is proportional
to the reference strain u, inversely proportional to
the square of the stability number, and is also
influenced by prop spacing ratio in a manner that
has not yet been included in a simple formula. It
has been demonstrated that large deformations
must be expected of braced walls in normally
consolidated clays, bringing the soil close to
failure. Stiffening the wall sufficiently to reduce
deformation is unlikely to be economic. The use of
soil stabilisation technology, to construct deep in
situ props prior to excavation, is more promising.
7. The design of props is best undertaken with
respect to existing rules based on previous
observations. Optimization is best understood in
relation to construction monitoring and design-asyou-go.
8. Excavations, especially in soft clays, are risky.
They should be monitored carefully during
construction. Setting a limit for the ratio wmax/H
observed during construction may be rational in
relation to three issues: the avoidance of serious
damage to the retaining wall itself; the avoidance
of serious damage to pre-existing structures or
services nearby; the ability to control the rate of
deformation by preventing soil from approaching
too close to its mobilization of peak strength.

Rational limiting values are suggested in relation

to the particular characteristics of the job. This
makes it difficult to regulate in terms of simple
fixed
limits,
although
procedures
are
recommended for engineers to follow.
9. It has been demonstrated that deep
excavations in soft clay pose an unusual
deformation problem. In the past, engineers have
relied on measuring soil strength and using
conventional safety factors. It has now been
demonstrated that the acquisition of soil stressstrain data is essential to predicting field
behaviour. Simple formulae then predict
deformations within a factor of 2.9 (two standard
deviations) within minutes. MSD, in a few hours
on a spreadsheet, can make more accurate
predictions of the staged deformation profile, both
wall displacements and subsidence trough, within
a factor of 1.4. It is unclear in simple cases
whether FEA would offer better predictions that
that: ground parameters especially remain
inherently difficult to assess. However, FEA using
a simple parabolic stress-strain relation for clays,
might be very valuable for analysing unusual
designs.
10.
There is a need for further work on
mechanisms associated with volume change,
such as for clay consolidation or for sands.
REFERENCES
Anderson, D. G. and Richart, F.E. (1976) Effect of
straining on shear modulus of clays. Journal of
the Geotechnical Engineering Division ASCE,
Vol. 102, No. GT9, pp. 975-987.
Batten M. and Powrie W. (2000) Comparison of
measured and calculated temporary prop loads
at Canada Water station, Geotechnique, Vol. 50,
No. 2, pp. 127-140.
Bjerrum, L., and Landva, A. (1966) Direct simpleshear test on a Norwegian quick clay.
Geotechnique, 16, No.1, 122.
Bolton, M.D., Lam, S.Y. and Osman, A.S. (2008)
Supporting excavations in clay - from analysis to
decision-making. 6th International Symp. on
Geotechnical
Aspects
of
Underground
Construction in Soft Ground, Taylor and Francis,
pp. 15-28.
Bolton, M.D. and Powrie, W. (1988) Behaviour of
diaphragm walls in clay prior to collapse,
Geotechnique, Vol. 38, No.2, pp. 167-189.
Boone S.J. (2006) Deep Excavations: General
Report. 5th International Symp. on Geotechnical
Aspects of Underground Construction in Soft
Ground, Amsterdam, pp. 23-31.
British Standards Institution (1994). Code of
practice for earth retaining structures. BS 8002.

Burland, J.B. and Wroth, C.P. (1974) Settlement

of
buildings
and
associated
damage.
Proceedings of Conference on Settlement of
Structures, Cambridge, Pentech Press, pp. 611654.
Burland, J.B., Broms, B.B. and de Mello, V.H.B.
(1977) Behaviour of foundations and structures.
Proceedings of 9th International Conference on
Soil Mechanics and Foundation Engineering,
Tokyo, Vol. 2, pp. 495-546.
Burland J.B., Standing J.R and Jardine F.M.
(2001) Building Response to Tunnelling: Case
Studies from Construction of the Jubilee Line
Extension, CIRIA Special Publication 200,
Thomas Telford.
Clarke P.J. and Prebaharan, N. (1987) Marina
Bay Station, Singapore, excavation in soft clay,
Proceedings of Case Histories in Soft Clay,
NTU, Singapore, pp.95-107.
Diaz-Rodriguez, J.A., Leroueil, S. and Aleman,
J.D. (1992).Yielding of Mexico City clay and
other natural clay. Journal Geotechnical Div.
ASCE, Vol.118, No.7, July, 1992.
Finno, R.J. and Chung, C.K. (1990). Stress strain
strength responses of compressible Chicago
Glacial clays. Journal Geotechnical Division
ASCE, Vol.118, No.10, pp. 1607-1625.
Gaba A.R., Simpson B., Powrie W. and Beadman
D.R. (2003)
Embedded retaining walls
guidance for economic design, CIRIA Report
C580, CIRIA, London.
Hunt, C. E., Pestana, J. M., Bray, J. D., and
Riemer, M. (2002). Effect of pile driving on static
and dynamic properties of soft clay. Journal
Geotechnical
and
Geoenvironmental
Engineering, Vol.128, No.1, pp.1324.
Jen, L.C. (1998). The design and performance of
deep excava-tions in clay. PhD thesis, Dept. of
Civil and Environmental Engineering, MIT.
Kim, T. C. and Novak, M. (1981) Dynamic
properties of some cohesive soils of Ontario.
Canadian Geotechnical Journal, Vol. 18, pp.
371-389.
Lam (2010) Ground movements due to deep
excavation: physical and analytical models. PhD
thesis, University of Cambridge.
Lam, S.Y. and Bolton, M.D. (2010) Energy
conservation
as
a
principle
underlying
mobilizable
strength
design
for
deep
excavations, ASCE Journal of Geotechnical and
Geoenvironmental Engineering, to be published.
Lin, H.D. and Wang, C.C. (1998). Stress strain
time function of clay. Journal Geotechnical and
Geoenvironmental Engineering, ASCE, Vol. 124,
No. 4, pp. 289-296.
Magnus, R., The, C.I. and Lau, J.M. (2005) Report
of the Committee of Inquiry into the incident at
the MRT Circle Line worksite that led to the
collapse of Nicoll Highway on 20 April 2005,
Singapore Government Ministry of Manpower.

Moh, Z.C., Nelson, J.D. and Brand, E.W. (1969).

Strength and Deformation behaviour of Bangkok
clay. ICSMFE VII, Mexico, 1969, pp.287-295.
ORourke, T.D. (1993) Base stability and ground
movement prediction for excavations in soft clay.
Retaining Structures, Thomas Telford, London,
pp. 131-139.
Osman, A.S. and Bolton, M.D. (2004) A new
design method for retaining walls in clay.
Canadian Geotechnical Journal, 41 (3), 451-466.
Osman, A.S. and Bolton, M.D. (2006) Ground
movement predictions for braced excavations in
undrained clay. ASCE Journal of Geotechnical
and Geo-environmental Engineering, 132 (4),
465-477.
Park R. and Gamble W.L. (2000) Reinforced
Concrete Slabs, John Wiley & Sons.
Richards, D.J. , Clark, J. and Powrie, W. (2006)
Installation effects of a bored pile wall in
overconsolidated clay, Geotechnique Volume
56, No. 6, pp. 411-425.
Shibuya, S. & Mitachi, T. (1994) Small strain
modulus of clay sedimentation in a state of
normal consolidation. Soils and Foundations,
Vol. 34, No.4, pp. 67-77.
Shirlaw, J.N., Tan T.S. and Wong, K.S. (2006)
Deep excavations in Singapore Marine Clay. 5th
International Symposium on Geotechnical
Aspects of Underground Construction in Soft
Ground, Amsterdam, pp. 1-16.
Stroud, M. A. (1974) The Standard Penetration
test in sensitive clays and soft rocks. Proc.
European seminar on penetration testing,
Stockholm, Vol. 2:2, pp. 366-375.
Tan, T.S. and Shirlaw, J.N. (2000) Braced
excavation Excavation in general, 3rd
International Symposium on Geotechnical
Aspects of Underground Construction in Soft
Ground, Tokyo, pp. 53-62.
Twine D. and Roscoe H. (1999) Temporary
propping of deep excavations guidance on
design, CIRIA Report C517.
Vardanega, P.J. and Bolton, M.D. (2010) The
stiffness of clays and silts, under review.
Vucetic M. and Tabata K. (2003) Influence of soil
type on the effect of strain rate on small-strain
cyclic shear modulus. Soils and Foundations,
Vol. 43, No.5, pp.161-173.
Whittle, A.J. (1993). Evaluation of a constitutive
model for overconsolidated clays. Geotechnique,
43, No.2, pp. 289-313.
Williams B.P. and Waite D. (1993) The design
and construction of sheet piled cofferdams,
Special Publication 95, CIRIA, London
Wong, K. S., Broms, B. B. (1989). Lateral wall
deflections of braced excavations in clay. ASCE
Journal of Geotechnical Engineering, Vol. 115,
No. 6, pp. 853-870.